Attribute reference design

[chamalulu]

Hello.
I think I'm aware of how attribute access is resolved in python. When
referencing a class instance attribute which is not defined in the
scope of the instance, Python looks for a class attribute with the
same name. (For assignment or deletion this is not the case,
thankfully.)
I've been trying to understand why? What is the reason behind, or
practical purpose of, this design decision? Anyone, please enlighten
me.

/Henrik

 

[Diez B. Roggisch]

Hello.
I think I'm aware of how attribute access is resolved in python. When
referencing a class instance attribute which is not defined in the
scope of the instance, Python looks for a class attribute with the
same name. (For assignment or deletion this is not the case,
thankfully.)
I've been trying to understand why? What is the reason behind, or
practical purpose of, this design decision? Anyone, please enlighten
me.

How else would you resolve methods which are of course defined on the
class but invoked through the instance?

Diez

 

[chamalulu]

How else would you resolve methods which are of course defined on the
class but invoked through the instance?

Yes, of course... You're right.
Didn't think of that.
Thank you. I'll go stand in the corner.

I think I haven't got this bound/unbound stuff through my head yet. If
I dir() a class instance I see the methods right there. Are they not
bound to the class instance at instanciation (and as such be
attributes of the class instance)?

/Henrik

 

[Gary Herron]

Yes, of course... You're right.
Didn't think of that.
Thank you. I'll go stand in the corner.

No need. Also, you can define a class attribute (C++ might call it a
static attribute) and access it transparently through an instance.

class C:
aClassAttribute = 123
def __init__(self, ...):
...

c = C()
.... do something with c.aClassAttribute ...

 

[chamalulu]

No need. Also, you can define a class attribute (C++ might call it a
static attribute) and access it transparently through an instance.

class C:
aClassAttribute = 123
def __init__(self, ...):
...

c = C()
... do something with c.aClassAttribute ...

Actually, this is why I started too look into the attribute reference
mechanics to begin with. Coming from mostly C# development I think
it's weird to be able to refer to class attributes (static members)
through a class instance (object). But I think I'm getting the
picture. Function objects lay flat in memory (some heap...). When
defined inside classes they are wrapped in method objects. When
refered through classes or class instances they are unbound method
objects or bound method objects respectively. Am I on the right track?
I still don't get why these methods show up when I dir() a class
instance.

/Henrik

 

[Gary Herron]

Actually, this is why I started too look into the attribute reference
mechanics to begin with. Coming from mostly C# development I think
it's weird to be able to refer to class attributes (static members)
through a class instance (object). But I think I'm getting the
picture. Function objects lay flat in memory (some heap...).

Not quite. Not *flat memory* or a heap. Modules are first class
objects, and functions (and classes and anything else) defined at the
top level of the module are *attributes* of the module.

Gary Herron

 

[Bruno Desthuilliers]

chamalulu a écrit :

Actually, this is why I started too look into the attribute reference
mechanics to begin with. Coming from mostly C# development I think
it's weird to be able to refer to class attributes (static members)
through a class instance (object). But I think I'm getting the
picture. Function objects lay flat in memory (some heap...).

Python's functions are ordinary objects, instance of type 'function'.

When
defined inside classes they are wrapped in method objects.

Nope. The wrapping happens at lookup time, thru the descriptor protocol
(the same thing that gives support for properties).

When
refered through classes or class instances they are unbound method
objects or bound method objects respectively.

That's what function.__get__() returns, yes. What is stored in the class
object's __dict__ is the plain function.

Am I on the right track?
I still don't get why these methods show up when I dir() a class
instance.

"""
Help on built-in function dir in module __builtin__:

dir(...)
dir([object]) -> list of strings

Return an alphabetized list of names comprising (some of) the
attributes
of the given object, and of attributes reachable from it:

No argument: the names in the current scope.
Module object: the module attributes.
Type or class object: its attributes, and recursively the
attributes of
its bases.
Otherwise: its attributes, its class's attributes, and recursively the
attributes of its class's base classes.

"""

 

[chamalulu]

Nope. The wrapping happens at lookup time, thru the descriptor protocol
(the same thing that gives support for properties).

Aha, I should read up on that.

Help on built-in function dir in module __builtin__:

So, the dir function is a little more helpful than I thaught. I
checked instance.__dict__ now instead of dir(instance). No methods.
It's getting clearer. Thank you.

I think you've provided me with a good reason for class instances to
delegate attribute references to the class if class instance doesn't
contain the attribute. Diez, Gary, Bruno; Thanks for your help.

/Henrik

 

[Cédric Lucantis]

Le Wednesday 02 July 2008 01:17:21 Gary Herron, vous avez écrit :

No need. Also, you can define a class attribute (C++ might call it a
static attribute) and access it transparently through an instance.

class C:
aClassAttribute = 123
def __init__(self, ...):
...

c = C()
... do something with c.aClassAttribute ...

Be very careful with that, as it looks like C++ or similar other OO languages,
but python handles it in a strange way if you assign a value to such an
attribute. It took me a long time to understand what happens here:

class Foo (object) :
bar = 0

foo = Foo()
print '(1) Foo.bar: %d' % Foo.bar
print '(1) foo.bar: %d' % foo.bar

Foo.bar += 1
print '(2) Foo.bar: %d' % Foo.bar
print '(2) foo.bar: %d' % foo.bar

foo.bar += 1
print '(3) Foo.bar: %d' % Foo.bar
print '(3) foo.bar: %d' % foo.bar

here's the output:

(1) Foo.bar: 0
(1) foo.bar: 0
(2) Foo.bar: 1
(2) foo.bar: 1
(3) Foo.bar: 1 # hey dude, where is my bar ?
(3) foo.bar: 2

In the third case, you might expect foo.bar += 1 to just increment Foo.bar,
but actually it doesn't. I first thought it was a bug, but it's not. When you
write foo.bar += 1 (equivalent to foo.bar = foo.bar + 1), python first looks
for a bar attribute, finds it in the class members, and then creates an
instance member with the result. Things would be different with a mutable
type implementing the += operator.

I discovered this in a middle of a project and it was hard to track all these
assignments in my code to correct them, so I'd suggest to always access class
members through the class instead of the instance, unless you have no choice
or know exactly what you are doing.